Source: Microsoft Foundry — Agent Service Overview

Foundry Agent Service: prompt, voice and hosted agents on the Responses API, toolboxes via MCP, Entra agent identity, VNet isolation and publishing.

Contents

Summary

Overview of the managed agent platform inside Foundry and the spectrum from declarative to full-code agents.

Key claims

  • Prompt agents: declarative (instructions + model + tools), fully hosted, versioned; optional Microsoft Copilot harness (preview) for multi-step tool and skill workflows.
  • Hosted agents: your container or .zip (Agent Framework, LangGraph, OpenAI Agents SDK, Anthropic Agent SDK or custom), run with a managed endpoint, autoscaling, a dedicated Entra identity and session state.
  • Voice-based prompt agents run over Voice Live, either native speech-to-speech or a cascaded STT → text model → TTS pipeline (preview).
  • Ephemeral agents: call the Responses API directly with the definition in your code.
  • Toolboxes bundle tools (web search, file search, code interpreter, MCP servers, functions, OpenAPI) behind one managed MCP endpoint with central auth and versioning.
  • MCP auth options: key, Entra managed identity, OAuth on-behalf-of passthrough, or none.
  • Private networking for prompt agents; hosted agents support BYO VNet with VM-isolated sessions. Standard setup uses your own Storage, AI Search and Cosmos DB for state.
  • Publishing to Teams, Microsoft 365 Copilot and the Entra Agent Registry; A2A v1.0 GA.

Feeds

Azure AI Foundry

Suggest changes →